NOTE ON FIBONACCI PRIMALITY TESTING John Brillhart University of Arizona, Tucson, AZ (Submitted August 1996-Final Revision January 1997)

Transcription

1 John Brillhart University of Arizona, Tucson, AZ (Submitted August 1996-Final Revision January 1997) 1. INTRODUCTION One of the most effective ways of proving an integer N is prime is to show first that N is a probable prime, i.e., that a N ~ l = 1 (mod N) for some base a l<a<n-l, then to find enough prime factors of N±l so that certain other conditions are satisfied (see [1] for details of such primality tests). The problem of finding these prime factors is, of course, the difficult time-consuming part of this process, anything that assists in the factoring of N±l is of great value, particularly when N is large. In the case of the Fibonacci Lucas numbers F n L n, we are quite fortunate that identities exist whose form is exactly suited to this purpose. (These were discovered by Jarden [4, pp ]. Their use in primality testing was first made by the author in the early 1960's see [4, p. 36].) The identities are all quite simple, asserting that F n ±\ L n ±\ are equal to a product of certain Fibonacci Lucas numbers with subscripts smaller than w, which numbers in turn may well have many known prime factors. Examples of these identities are: F 4 k+i - 1 = Fkh^ik+i a n d ^ 4 * + i + 1 = F ik+\hk With the assistance of this set of identities, many large i^'s L n % s have been identified as primes [2, p. 255]. In this note we give a collection of similar, but more complicated identities that can be used to establish the primality of'the primitive part F of F k, i.e., the cofactor remaining after the algebraic factors of F k have been divided out. This cofactor is given by the formula (see [2, p. 252]) F* = JjFf k/d \ ju the Mobius function. (1) d\k The subscript of F^ in the identities in the present collection has at most two distinct prime divisors, since an identity with three or more prime factors does not in general have a simple multiplicative structure on its right side, i.e., the right side is not just a ratio of products of F k ' Z^s. The case of two prime divisors is transitional in that some identities have simple multiplicative structure others do not [see (17) (18)]. 2, THE IDENTITIES In the proofs that follow, we use elementary Fibonacci Lucas identities. Also, throughout this note we use the familiar identity F 2r - F r L r without further mention. In the first two theorems, the subscript of F is a power of a single prime. Theorem 1: For n > 3, 222 [JUNE-JULY

2 K n-2 F r +1 = ^CL - ( 3 ) r 2"~ 2 Proof of (2): Substituting r = 2 n ~ l s = 2 n ~ 2 into the identity we obtain L r L s = L r+s+ (-l) s L r _ s, (4) r 2 n ~ l 2"~ 2 Proof of (3): Making the same substitution into the identity F s L r =F r+s -(-iyf r _ s, (5) we obtain Theorem 2: Let p = s (mod 4) be a prime, where s - ±1. Then, for n > 1, 2 n ~ 2 F* n _ l = p n -\ps)l2 L p n -\p+s)l2,gv V' J7*, j =!f- l (p+s)/2 p n -\p-s)l2 (j, P n F, Proof of (6): If we substitute r = i? " - 1 ^ ) 5 = p " " 1 ^ ) into the identity use the fact that F n = F n for n odd, then we obtain F r+s = F r L s -(-iyf r _ s, (8) F* = p " = p n ~ l (p- y 2 P"~ 1 (P+S)/2, Proof of (7): This follows in the same way by setting r = p"" 1!^) j = ^""H^-)- Q i. For p = 3, formulas (6) (7) have a particularly nice form: i $ - l = 4,-, F 3 :+1 = Z 2. 3 _ 1. (9) 2. For p = 5, formulas (6) (7) are of not interest here, since F, n>2, has 5 as an intrinsic factor [2, p. 252] cannot be a prime. The numbers F* n /5 are dealt with in (26). J. For p = 7, formula (6) becomes the interesting formula r^n 1 = Lrj n \h^n-\-liyrjn-\- (10) 4 In general, if JV = -j (i^* ± 1) is a probable prime, then JV +1 = \ (F +1). 1998] 223

3 In the next theorems, the subscript of F has two different prime factors. Theorem 3: Let q be an odd prime, then for n > 1, F* n - (_l)(<?-l)/2 = q n -\q+l)l2 r q n -\q-l)l2, j jv F* + ( - 1 ) ^ - ^ 2 - g" \q+l)/2 q"! (g-l)/2 s^) Also, for m > 2, we have r * _ 1-2 m - l q n -\q+\)l2 r 2 m - l q n -\q-\)i2 (, ~x r 2 m q" l T V 1 J ' 2 q L^m-Xf-X Proof of(11): Substituting r = q"- l {^) s = ct\3±) mt0 L r+s = 5F r F s + {-\) s L r _ s, we obtain F* = F 2q n q n - X q n 5F q»- l (q + l)/2 F q n - l (q-l)/2, n(q-l)/2 qn 2gn qn gn Proof of (12): Making the same substitutions as in (11) into (4) leads to F* = q" = L q n -\q+l)/2 L g n -\q-l)/2, r^q-y)l2 2q n T T V V V 1 V 1 Proof of (13) (14): These results are obtained similarly by using r = 2 m ~ l q n ~ l (2±L) s = 2 m - l q"- l { ±) as in (11) (12). D Theorem 4: If p<q,p q odd primes, then for m, n > 1, ce * _, //"-y- 1 V - y - 1 (q-lfp m - l q n - 1 (<?+!) 1 p m q n ~ l (15) A ( - 1 ) ^ 5 r />-, > ^ y - i r / - ' g - Proof: For brevity's sake, put w = p n ~ l q n ~ l. Then, using the formula (see [5, p. 209, (79)], P-\ F P n = i(-v- -( P ; r )^- r Fr 2r, n odd, (16) we have that 224 [JUNE-JULY

4 FF A w x pqw -F F qw* pw = ^±(-lx^(^- r )5^/r^_^eiX^P-rj 5^^ra P ~ 7 - I Y ^ ^ i n r ^ p-3 S,... /, (p-r\^-r\f^l- 2r -Fr l - 2r ^sf^f^i-f^ti-iy^f^;^ r=q F-' V " ' I F qw~ F w But, using the identity F^ - (-l)* (m 1} i^2 = i ^ + i ^ ^ - i ) w i t h k = w nt = q,we have Thus, ^w^pqw ~~ Fqvrpw C I Jr qw r w r w(q-l) r w(q+l) l-^'r'h^^f}- J7* 1 pqw^w gw pw P pqw ~ l ~ 17 F -* qw x pw -1 < p-\-2r z7/?-l-2r > _ 5F w F w(q-l) F w(q + l) V / i y P (p-r\^-r\ F q P w * ~ F w" [ It is worth while to give some special cases. Corollary 5: If q is a prime, then for m, n > 1, ^/7 J7 /7 for q>7, P 3 m g n-l 9 C / J7* (g" 1 ) 5 9 (g +1 ) /z?2. J7 2 A n $ n ^5m 5m^«fl «- I - ^ l^v-l^h \^ 5m-l «-T- + P J- 5m-\ 5m-\ a n-\ n * c m _ n - l ~ *J (1 ) r 5«V The following are some further simple cases. Here q is a prime. ^ - 1 = 1 ^ - 1 ^, ^ 5, (19) ^ - 1 = 5 / ^ / ^, ^ 7, (20) / v ; - l = ^/^_ 1 /V +1 (25F/-10F? 2 +4), qr^ll. (21) The next is a formula containing a "+1 M. From numerical evidence, there seem to be few identities with a "+ H that have a right side with a multiplicative structure. 1998] 225

5 Theorem 6: If q > 5 is a prime, then ^ ; + i = ^ - (22) Proof: Using F 3r = F r (5F? +(-l) r 3) Zg r = Z r (5F r 2 +(-l) r ), we find that Z g (F 39 +2F q ) = L q F q (5F q 2-3)+2F q L q = F q L q (5F*-l) = F q I^q. Thus, /Some Examples: We consider the factorizations leading to proofs of the primality of the probable primes F* 45, i^* 2853 F* In the first, we have F* 45 = p; 29 = -p0- = Then, by (20) Tables 2 3 in [2], we find the complete factorization: In the second, identity (20) gives ^5*29 ~ 1 = 5^28^9^30 = 5 ( A * M V X ^ 9 ( A 5 ^ ) = 5( )( )(2 2 -ll ). ^*285 ~ 1 = ^5*457 ~ 1 = 5^456^457^458 = ^(^228M 14^57^7)^457(^229^29) 5 each factor of which is again completely factored using the tables in [2]. The primality of F* 45 ^2285 ls established, respectively, from these complete factorizations using Theorem 1 in [1]. In the third, identity (21) is used to obtain A 4203 ~~ 1 = ^ ~^ = TJ ^2028^030^ = IT ^ ^ )(A ) ' where G = 25i^Q 29-10i^Q As it happens, all the F k 's L k 's can be factored completely G is partially factored as G = c, where c is a 1682-digit composite cofactor. Since the logarithm of the product of the 64 known prime factors in these factorizations (counting multiplicity) is about 33.9% of the 2544-digit number F* 4203, the "cube root" Theorem 5 in [1] can be used to establish the primality of this number. Fourteen of these factors have more than 20 digits. For another example, see [3, 4], where (18) is used in the primality proof of the 1137-digit probable prime i^*225- A final example is the probable prime ^49, for which not enough prime factors have been discovered to complete a primality proof. I would like to thank W. Keller for suggesting the above examples for sending me information about them. The next theorem deals with those F*'s that have an intrinsic factor, which is divided out of the primitive part. Only the first power of an intrinsic factor can divide the primitive part. 226 [JUNE-JULY

6 Theorem 7: We have % + l = fl^ ^ 2, (24) % + l = ^ n ^ l, (25) ^ - 1 = 5 / ^ / 3., ^, /1 > 2, (26) ^ + 1 = ^ ^, ( 4 ^, - 7 ^ ),» > 1. (27) Proof of (23) (24): Using Zg r = Z r (Ar - (-1/) 4 r = Z? - (-l) r 2, we have Now, firom Z^-(-l) r 5 = 4-i4+i> w e h a v e -^3*2" ~ 2 = : ^ 2 M - 1 _ 5 = ^^-A^+p from w h i ch t h e identity follows. Also, from the equalities in (28), we have \^KT +2) = (Z 2 +l) = i(z^_ 1-1) = ^(Z 2 _, -1)(Z 2 _, +1) ^ ^ fc=2 t=2 k=\ Proof of (25): Using L,,. = L r (L; - (-l) r 3), we find that which implies the result. 77 J7 T 77* ^ " ^ r 1 = IT = T2 _ o 4-3" 17 F r ^2-S"- 1 ' J 2-3" 4-3"- 1 ^Z.3"- 1 Proof of(26): From (16), we obtain the formula F 5r = 5F r (5F r 4-5F r 2 +1), so % - 1 = ^f 1 = 5 2 ^ ( /, -1) = SF^F^F^, using F? + {-W=F r -ifr+x- Proof of (27): From [4, p. 212, (86)], ^ (-l)^v)^«odd, so L, = Z (4-1L\ + ULI-7). Thus, 1998] 227

7 from which the identity follows. D F* = F * rf * r - 1 = i ^ - = L 6., _, - 1L\ _, + \4L] _, " /7 C 7" 4-7"" 1 ' ^ 4-7 n ' n-\ '4-7 n ' Remark: Numerical evidence suggests that, for n > 2, there are no multiplicative formulas for the even integers N x = (F* 3 13) - 1 N 2 - (F* n / 5) +1. On the other h, if ±N X or y N 2 should be probable primes, then the following formulas, which relate back to (24) (25), might be useful in establishing their primality: 2 l 2 (K* Fl \ 4-3" + 1 J ~N 7 -l = There are some other formulas involving F* L* of various kinds, but these will not be considered here. We conclude this note by observing that the identities used in the proofs, such as those in (4), (5), (8), each contain the factor (-1)*, which becomes the ±1 in the identity for F*±l. In general Lucas sequences, of which the pair {F n } =0 {L n } =0 is a special case, this factor is Q s. Thus, the other Lucas sequences that have formulas like those in this note are those for which g = l(see[l,p.627]). REFERENCES 1. J. Brillhart, D. H. Lehmer, & J. L. Selfridge. "New Primality Criteria Factorizations of 2 w ±l. n Math. ofcomp (1975): J. Brillhart, P. L. Montgomery, & R. D. Silverman. "Tables of Fibonacci Lucas Factorizations." Math. ofcomp (1988):251-60; S1-S H. Dubner & W. Keller. "New Fibonacci Lucas Primes." Math. ofcomp., to appear. 4. D. Jarden. Recurring Sequences. 3rd ed. Jerusalem: Riveon Lematematika, E. Lucas. "Theorie des fonctions numeriques simplement periodiques." Amer. J. Math. 1.1 (1878): ; AMS Classification Numbers: 11A51, 11B39 F* 5 ~-l [JUNE-JULY